Re: [Discuss-gnuradio] thread safety of gr_wavfile_sink::close()

[Eric Blossom]

On Fri, Apr 30, 2010 at 08:48:48PM -0400, Don Ward wrote:
> 
> ----- Original Message ----- From: "Eric Blossom" <address@hidden>
> To: "Don Ward" <address@hidden>
> Cc: "gnuradio mailing list" <address@hidden>
> Sent: Friday, April 30, 2010 8:25 PM
> Subject: Re: [Discuss-gnuradio] thread safety of gr_wavfile_sink::close()
> 
> 
> >On Fri, Apr 30, 2010 at 07:42:32PM -0400, Don Ward wrote:
> >>I have a question about the thread safety of
> >>gr_wavfile_sink::close().  The description says it is thread safe
> >>and it uses d_mutex, but what is to stop the work() function from
> >>writing the file at the same time that close() (via close_wav()) is
> >>writing and closing the file?  gr_wavfile_sink::work() does not lock
> >>d_mutex (except when d_updated is set).  Is there something going on
> >>behind the scenes to prevent the threads from mangling the output
> >>file, or is this a problem with the code?
> >>
> >>The reason I am interested is because I would like to do something
> >>similar with gr_udp_sink, allowing one socket to be closed and a new
> >>one to be created when one client finishes and another client starts
> >>up.
> >>
> >>Thanks,
> >>
> >>-- Don W.
> >
> >It looks fairly hosed up.  One fundamental design/documentation
> >problem is that the header file does not document which variables must
> >be protected by the mutex.  Also, proper use of the mutex (including
> >in work) would eliminate all of the code related to d_updated.
> 
> So, it is ok to use the mutex across the work function?

Yes, as long as nobody who's holding it ever blocks.

> >While I'm at it, set_bits_per_sample quietly fails for any value not
> >equal to 8 or 16.
> >
> >gr_udp_sink has similar problems.
> 
> Ok.  Let me work on gr_udp_sink, then you can tell me how close I
> came to getting it right.
> 
> -- Don W.

Thanks!
Eric